KR102273720B1 - Bill discrimination apparatus - Google Patents

Abstract

The present invention relates to a bill discriminating device that can be easily and quickly applied to preparing learning data and discriminating bills by minimizing an increase in the amount of learning even when adding a banknote type and country.

Description

Bill discriminating device {BILL DISCRIMINATION APPARATUS}

The present invention relates to a banknote discrimination device, and more particularly, to a banknote discrimination device that can be applied easily and quickly to prepare learning data and to differentiate banknotes by minimizing an increase in the amount of learning even when adding a banknote type and country.

Institutions, organizations, and companies that process large amounts of money such as banks, especially banknotes, use a counter to count the number of banknotes. In general, counters are divided into bills and coins.

In the case of banknote counters, in the past, only a function of determining the quantity of banknotes was used, but recently, an integrated function has been improved and used as a device to distinguish between counterfeit money and denomination types.

To this end, it is common for recent banknote counters to acquire various information such as visible light, ultraviolet light, infrared light, and magnetism during the counting process of banknotes and process them to analyze the type of bill, whether counterfeit money, and quantity.

In order to make this processing more efficient and accurate, a device to which a deep learning (or machine learning) technique is applied has recently been developed and used on acquired data. The device using such deep learning is a method of classifying counterfeit money and bills by pre-learning the type of ticket to be distinguished and discriminating by the learning data.

By applying the deep learning method, the range of response to new counterfeiting techniques could be increased and the accuracy could be improved, but problems with prior learning are being raised.

For example, since banknote discrimination is performed through one learning data, if a country or denomination for identification is added, re-learning of all denominations is required, and as the number of denominations and countries increases, the time and effort required for learning There are problems that increase exponentially.

Republic of Korea Patent No. 10-1965751 (Registration date: March 29, 2019) Apparatus and method for multi-country classification classification based on convolutional neural network

Accordingly, an object of the present invention is to solve this problem, and to provide a banknote discrimination device that can be easily and quickly applied to preparing learning data and identifying banknotes by minimizing the increase in the amount of learning even when adding a banknote type and country.

In order to achieve the above object, a banknote discrimination apparatus according to the present invention includes: a banknote data acquisition unit for generating banknote data of a banknote to be discriminated; a calculation unit for discriminating the banknote data using a pre-learned first algorithm to discriminate the banknote data, and discriminating the banknote data in which the area is determined by the pre-learned second algorithm to determine the type of bill; and a storage unit in which the first algorithm and the second algorithm are stored.

The banknote discrimination apparatus according to the present invention minimizes the increase in the amount of learning even when adding the banknote type and national stamp, so that it can be easily and quickly applied to prepare learning data and to distinguish banknotes.

1 is an exemplary diagram schematically illustrating the configuration of a banknote discrimination device according to the present invention.
2 is an exemplary diagram for explaining the learning of the banknote identification algorithm and the identification process using the same.
3 is an exemplary view for explaining a discrimination system through a banknote discrimination algorithm.
Figure 4 is an exemplary view for explaining a banknote discrimination method according to the present invention.

Hereinafter, embodiments of the present invention will be described so that those of ordinary skill in the art can easily implement them with reference to the accompanying drawings. In the accompanying drawings, it should be noted that the same reference numbers are used as much as possible when indicating the same configuration in other drawings. In addition, in the description of the present invention, if it is determined that a detailed description of a related known function or a known configuration may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted. In addition, certain features presented in the drawings are enlarged, reduced, or simplified for ease of description, and the drawings and components thereof are not necessarily drawn to scale. However, those skilled in the art will readily appreciate these details.

1 is an exemplary diagram schematically illustrating the configuration of a banknote discrimination device according to the present invention.

Referring to FIG. 1 , the banknote discrimination device according to the present invention may include a driving unit 10 , a storage unit 20 , a banknote data acquisition unit 30 , an operation unit 40 , and a result output unit 50 . have.

For the discrimination device described below, the configuration of the initial learning device of deep learning will not be separately described, and the description will be focused on the discrimination device. In the case of the learning device, since it can be implemented in a configuration similar to that of the discrimination device or can be configured in a variety of forms, such as in an environment using a computer, the description of the learning device will proceed without limiting the specific form. Therefore, in the following, the driving of the discrimination device can be interpreted as driving the learning device. For example, the description of 'discrimination' is used as the term 'learning', except when it is necessary to distinguish between 'distinguishing' and 'learning'. can be interpreted and replaced.

The driving unit 10 serves to move the bill that needs to be discriminated to a designated position. That is, it is provided to move the discriminated bills in the loading box of the discrimination device to the location where the sensors of the bill data acquisition unit 30 are located or to the outlet. The driving unit 10 may be configured with devices such as a motor, a gear, a wheel, and a roller, and performs operations related to separating the inserted banknotes into sheets and transferring and discharging the separated banknotes. The driving unit 10 may be configured to include a sensor device different from the sensor device included in the banknote data acquisition unit 30 in order to detect the movement position of the banknote and whether the banknote is operating normally. For example, by providing an optical sensor, it is possible to detect matters such as movement, position, and inclination of a bill, and a plurality of various sensors may be provided in addition. The driving unit 10 operates under the control of the calculating unit 40 .

The storage unit 20 stores the learning data and the discrimination algorithm to which the learning data can be applied, and is provided according to the request of the operation unit 40 . The discrimination apparatus of the present invention includes a discrimination algorithm to which a plurality of learning data is applied. Hereinafter, the discrimination algorithm will be defined as including the corresponding learning data and will be explained.

Specifically, the discrimination algorithm is composed of a first algorithm for classifying a region or country of use, and a second algorithm for classifying a class by region or country. In particular, the second algorithm is provided for each region or country to be discriminated, and a plurality of algorithms are provided. That is, the region of use of the currency is divided by the first algorithm, such as whether it is Australian dollar, US dollar, or British pound. And by the second algorithm, items such as one dollar bill or ten dollar bill of the currency are distinguished. In particular, the second algorithm may include an algorithm for discriminating whether the banknote is counterfeit or damaged.

In this discrimination algorithm, learning data is prepared through learning in advance, and the learning data is applied to complete the algorithm. In particular, in the event of a change in the title type, such as the addition of a new certificate, the second algorithm and the first algorithm corresponding to the region are re-learned to prepare an algorithm for discrimination of the new certificate.

The banknote data acquisition unit 30 acquires data of a banknote to be discriminated, and transmits the acquired data to the calculation unit 40 . The data acquired by the banknote data acquisition unit 30 may acquire various types of data such as image data and magnetic data, and for this purpose, a sensor and a photographing device for acquiring each data may be provided. In particular, in order to apply the discrimination algorithm of the present invention, image data photographed by any one or more of visible light, infrared light, and ultraviolet light is obtained by the banknote data obtaining unit 30 and transmitted to the calculating unit 40 .

The calculation unit 40 applies the banknote data transmitted by the banknote data acquisition unit 30 to the discrimination algorithm to classify the region and the type of banknote. Specifically, when the bill data is transmitted, the calculator 40 applies it to the first algorithm to determine the region. And the calculating unit 40 determines the type of bill by inputting the bill data into the second algorithm corresponding to the determined region. In this case, the calculating unit 40 may perform counterfeiting determination, banknote damage determination, and direction determination using the second algorithm. This will be described in more detail below. In addition, the calculation unit 40 outputs the determination result through the result output unit 50 and controls the driving unit 10 to move the bill. For example, the calculating unit 40 may transfer the banknote determined through the driving unit 10 to the discharge position or return it to the loading position, and may move counterfeit or damaged banknotes to a separate loading location. The operation of the driving unit 10 may vary depending on the function, shape, and specific configuration of the driving unit 10 , and the present invention is not limited to any particular form.

The result output unit 50 outputs and provides the determination result transmitted from the operation unit 40 . The result output unit 50 may output the discrimination result through a sound output unit or a display device provided in the discrimination device, store the result in data form, or output the result in the form of transmitting the discrimination result to a separate terminal device. You may. To this end, the result output unit 50 may include a driving circuit for output to a sound output unit or a display device, or may include an interface or communication module for storing or transmitting data in the form of data.

2 is an exemplary diagram for explaining the learning of the banknote discrimination algorithm and a differentiation process using the same, and FIG. 3 is an exemplary diagram for explaining the identification system through the banknote identification algorithm.

2 and 3, the banknote discrimination of the present invention is performed over two steps. First, the region of the bill is divided, and a process of specifically discriminating the type of bill according to the divided region is performed. For this reason, the learning process is also conducted by region and each book type.

This two-step identification method has many advantages over the existing regional and breed identification methods using deep learning. As the number of regions and species to be specifically discriminated increases, the learning and discrimination time is reduced compared to the existing method. In addition, the time required for learning can be reduced compared to the existing method. In particular, when the addition of the priesthood occurs, the re-learning time is dramatically reduced.

Specifically, in order to discriminate the domain and the type at once, the learning time increases exponentially according to the number of types. Moreover, if the addition of a new book occurs, the existing data is ignored and re-learning is required for all the books from the beginning, which greatly increases the learning time, and it is difficult to quickly add a new book.

On the other hand, in the case of the present invention, it is possible to reduce the learning time by re-learning only the second algorithm of the region and the first algorithm for distinguishing the region from among the plurality of classifications of the region to which the added species belongs. In particular, by dividing the first algorithm and the plurality of second algorithms, the learning content of each algorithm is reduced, so that the time required for learning can be further reduced compared to the existing ones.

For this learning, the second algorithm is prepared for each region, that is, for each region that uses the same currency, such as the United States, Australia, Europe, and Korea. And, the first algorithm is provided to determine which region the bill belongs to.

In this case, the second algorithm is learned by using all banknotes for each region as learning data, and it is possible to apply conventionally known learning methods and discrimination methods. In particular, as in FIG. 3 , the second algorithm can also determine whether or not counterfeit money, damage, and direction of banknotes as in the existing banknote discrimination method using deep learning. To this end, the banknote data is pre-processed and processed in a form that can be processed by the second algorithm, and the details of the banknote are discriminated using this.

On the other hand, in the present invention, the first algorithm is used only for the purpose of classifying regions. Through this, it minimizes the delay that may be caused by configuring the deep learning algorithm in two steps and enables fast processing.

To this end, the first algorithm is learned using data as shown in FIG. 2 . Specifically, by understanding the learning of the first algorithm, integrated data of banknotes is prepared for each region. The integrated data of banknotes is obtained by extracting the characteristics of banknotes of each region, and may be geometrical features or the composition or arrangement of specific elements.

Specifically, taking the front side of a Korean banknote as an example, it can be understood as in the form shown in (a) and (b) of FIG. 2 . Figure 2 (a) shows the front side of the 10,000 won bill, (b) shows the front side of the 50,000 won bill.

Looking at the form of Korean banknotes (60: 60a, 60b), the portrait of a great man (62: 62a) along with the numerical value display (61:61a, 61b, 65: 65a, 65b) and the Hangul amount display (63: 63a, 63b) , 62b) are shown.

Although the numerical value indications 61 and 65, the Hangul price indication 63, and the portrait of a great person 62 have different exact positions and sizes for each bill type, the approximate positions of the bills 60 appear similar. In particular, since there is a difference in bills for each type of bill, when the sizes of bills are matched, these positions are almost in the same place.

Using these features, the integrated data is processed so that the geometrical features stand out by overlapping and simplifying each type. Alternatively, the feature point, that is, the numerical amount display 61, the Hangul amount display 63, and the distance and angle between the portrait of a great person 62 is simplified to a state in which it is possible to check.

The integrated data prepared in this way can be similar to (c). That is, only the features of the portion where the numerical value display 61 , the Hangul amount display 63 , and the portrait 62 of a great person exist are left by overlapping. At this time, each feature point is extracted in the form of a region by overlapping, so that detailed content is not described. That is, the figure 62c for the portrait of a great person, the numerical amount display areas 61c and 65c, and the character amount display area 63c are displayed.

In this process, the mark that appears only for each type of book disappears. For example, patterns and shapes assigned only to specific bill types, such as the strip pattern 64a at the bottom of the 10,000 won bill 60a and the mark 64b at the bottom of the 50,000 won bill 60b, are omitted, and only common shapes and arrangements are omitted. It remains in the integrated data.

Learning is carried out by using the correct answer data of deep learning, and learning is performed so that banknote analysis in the form of (c) is performed by the first algorithm. For this purpose, integrated data created by integrating banknotes for each region is prepared.

In this form, data processing can be simplified by extracting feature points from banknotes to part the region, and it becomes possible to reduce the time delay due to the application of the first-order algorithm.

Here, the data for the creation of the integrated data has been described by taking the visible light image of the banknote as an example, but the present invention is not limited thereto. That is, any of the infrared image, the ultraviolet image, or the magnetic image may be used if the composition and arrangement of the feature points of the banknote can be confirmed as described above. It is also possible to write

4 is an exemplary view for explaining a banknote discrimination method according to the present invention.

4, the banknote discrimination method according to the present invention includes a learning step (S10), a banknote data acquisition step (S20), a first algorithm application step (S30), a second algorithm application step (S40), and a result output step ( S50) may be included.

The learning step (S10) is a step of preparing the first algorithm and the second algorithm by learning. In this learning step (S10), integrated data for banknotes for each region is created, and a first algorithm is learned to classify regions by the integrated data. Then, a second algorithm is provided for each region, and learning of banknotes for each region is performed.

To this end, the learning step (S10) may include an integrated data preparation step (S1), a first algorithm learning step (S3), a second algorithm learning step (S5), and a new certificate checking step (S7).

The integrated data preparation step (S1) is a step of generating integrated data for each region's currency, and extracting features of the regional currency. That is, in the integrated data preparation step (S1), each integrated data is prepared by extracting the features of money for each region. In this integrated data preparation step (S1), when a new certificate is confirmed in the new certificate checking step (S7), the integrated data for each region currency including the new certificate is regenerated. That is, if the US dollar is newly issued, the integrated data for the US dollar is regenerated.

The first algorithm learning step (S3) is a step in which the first algorithm is repeatedly processed to classify regions by the integrated data, and learning is performed through this process. When the new certificate is confirmed, the first algorithm learning step (S3) is repeated, and the learning of the first algorithm is performed again using the existing integrated data for each region and new integrated data of the region including the new certificate. That is, when a new note in US dollars is issued and the integrated data is regenerated, in the first algorithm learning step (S3), the learning process is performed again using the previously generated integrated data of other regions and the newly created US integrated data. do.

The second algorithm learning step (S5) is a step in which learning proceeds by repeatedly performing the discrimination processing of banknotes for each region based on banknote data. In the second algorithm learning step (S5), if the new certificate is confirmed, re-learning is performed for the second algorithm in the area to which the new certificate is added. That is, in the case in which the American priesthood is added to the above-described example, the second algorithm for the United States is relearned.

The step of confirming the existence of a new certificate (S7) is a step of determining whether a new type of currency has appeared in a specific region. When a new bill is confirmed, the integrated data creation step (S1) and subsequent steps are performed again, and when a new bill is not confirmed, the first and second algorithms are applied and used for actual bill identification.

The banknote data acquisition step (S20) is a step of acquiring banknote data using the banknote data acquisition unit 30 for a banknote as a discrimination target. In the present invention, although the description is based on the visible light image, as described above, when it is possible to extract a feature point with high similarity from a plurality of paper types among the data acquisition means used for discrimination on banknotes, images or data other than visible light are used. It can be obtained and used.

The first algorithm application step (S30) is a step in which the calculating unit 40 applies the first algorithm to the banknote data obtained in the banknote data acquisition step (S20) to confirm the region of the banknote. That is, it is discriminated in the first algorithm application step ( S30 ) in which region the bill to be discriminated is used. Then, the operation unit 40 calls the second algorithm for discriminating the type of the book in the corresponding region according to the discrimination result. Here, when a plurality of bills are input as a discrimination target, the first algorithm is applied to the first several bills, or the first algorithm is applied to one bill per several bills at a certain interval, that is, the region can be discriminated. can

The second algorithm application step (S40) is a step in which the calculating unit 40 checks the region determined in the first algorithm application step (S30), and calls the second algorithm of the region to discriminate the type of bill and other matters. . Here, in the step of applying the second algorithm ( S40 ), the calculating unit 40 may determine only the type of bill through the second algorithm, or may further determine additional matters such as whether counterfeit money, the direction of the bill, and whether it is damaged.

The result output step S50 is a step of outputting the banknote discrimination result through the calculation unit 40 and the result output unit 50 . In this result output step (S50), the operation unit 40 may process data or change the output form to suit the device for outputting the result, and may include a process of controlling the driving unit 10 according to the result. have.

In the above, it has been shown and described as a specific example to illustrate the technical idea of the present invention, but the present invention is not limited to the same configuration and operation as the specific embodiment as described above, and various modifications are within the limits that do not depart from the scope of the present invention. can be carried out. Accordingly, such modifications should be considered to fall within the scope of the present invention, and the scope of the present invention should be determined by the following claims.

10: driving unit 20: storage unit
30: bill data acquisition unit 40: calculation unit
50: result output unit 60: banknote
61, 65: numeric value display 62: portrait of a great man
63: Hangul amount display 64: Pattern

Claims (5)

a banknote data acquisition unit for generating banknote data of banknotes to be discriminated;
The banknote data is discriminated by a first algorithm set to classify the zone of use or country to determine the zone of use of the banknote, and the banknote data in which the zone is determined is determined by a second algorithm set to classify the type of banknote by country a calculation unit to determine the type of vol.; and
A banknote discrimination device comprising a; a storage unit for storing the first algorithm and the second algorithm.
The method of claim 1,
The second algorithm is
Banknote discrimination device, characterized in that provided for each region. The method of claim 1,
The first algorithm is a bill discrimination device, characterized in that it is an algorithm set to classify the region or country of use based on the learning data extracted geometric features of the bills for each region. delete The method of claim 1,
the calculation unit
Banknote discrimination device, characterized in that the second algorithm further discriminates at least one of counterfeit banknotes, banknotes damage, and the orientation of the banknotes.

Images